You are on page 1of 11

See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/221226029

Robust Measurement of the Blocking Artefact

Conference Paper in Proceedings of SPIE - The International Society for Optical Engineering · February 2009
DOI: 10.1117/12.805726 · Source: DBLP

CITATIONS READS

2 35

2 authors, including:

Giovanni (Gianni) Ramponi


University of Trieste
178 PUBLICATIONS 3,450 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Visual Privacy View project

High Dynamic Range (HDR) image processing View project

All content following this page was uploaded by Giovanni (Gianni) Ramponi on 15 July 2014.

The user has requested enhancement of the downloaded file.


Robust Measurement of the Blocking Artefact
Giovanni Ramponi and Leonardo Abate
IPL, DEEI, Università degli Studi di Trieste, Italy

ABSTRACT
A method is presented to measure the intensity of the blocking artefact in compressed pictures or video frames.
First, a way is devised to artificially introduce pure blocking, which closely resembles the real one subsequent
to JPEG compression. Then a modified no-reference measurement is proposed that requires less computations
than other formerly presented methods, permits to take into account the whole image or frame area, and is
not affected by interlaced video. Some first experiments indicate that the measured values relate closely to the
introduced blockiness effect. The robustness of the metric to the influence of other typical JPEG artefacts is
also checked. Further, the effect on blockiness of some enhancement strategies is measured. Pictures enhanced
with methods introducing the most severe blockiness are found to have the highest value of the proposed metric.
Finally the problem of blockiness measurement in video sequences is addressed. In this case the blocking grid is
no longer regular. In fact, blocks of different size could be used in encoding, and single blocks could be shifted
in referenced (P and B) frames due to motion compensation. A method is devised for grid detection.
Keywords: Blockiness, Still and video pictures, Artefact generation, Grid detection

1. INTRODUCTION
Blockiness in coded images and video is quite an annoying artefact, for at least two main reasons: it is very
visible even if it has moderate objective strength, especially in uniform areas of the picture; and, notwithstanding
the large amount of techniques which have been proposed, its cancellation is not yet a solved problem, due to
the negative effects on image sharpness which often result.
We focus our attention here on techniques which can be used to detect and quantify the presence of blockiness
artefacts, with metrics that take into account the response of the human visual system (HVS) in order to yield
significant results. We deal with still images at this stage; further developments will be devoted to single frames
of video sequences; temporal effects will be considered in a future work.
Some blockiness artefact detection algorithms that may be found in the literature are able to integrate an
overall video quality metric with a blocking dominant region segmentation, e.g.1 Most, however, are used with
the purpose of either specifically find the location of annoying blocking artefacts in an image or generically
determine the amount of blockiness disturbance present in the data. In the former case many of the most
recent operators act directly in the compressed, transform domain;2–4 some, which may be a part of a sharpness
enhancement method,5 are located after image decoding and work in the data domain. We are interested here
in algorithms whose output is an overall figure of merit for the image, as far as blockiness is concerned. For
example, methods of this kind are those in,6 in the frequency domain, or,7 in the data domain. For our purposes,
we specifically refer to the method suggested in,8 modified and improved in.9 We first deal with methods to
generate blocking on an image, in order to make some test images available for the evaluation of the performance
of different blockiness measurement algorithms, and we suggest a simple approach. Then, we propose a metric for
the detection and measurement of the blockiness artefact, which includes sub-operators that make its response
closer to the one of the HVS and, at the same time, more robust to other disturbances. Further, the metric is
used as a tool to be inserted in an image visualization chain in order to understand if a typical operator that
is applied to the image, i.e. detail sharpening, is compatible with the amount of image blockiness. Practical
examples demonstrate the effectiveness of the novel contributions that are provided.
In the second part of our work we deal with blockiness detection and evaluation in video sequences. In this case
Further author information:
G.R.: E-mail: ramponi@units.it,
L.A.: E-mail: leonardo.abate@deei.units.it

Image Processing: Algorithms and Systems VII, edited by Jaakko T. Astola, Karen O. Egiazarian
Nasser M. Nasrabadi, Syed A. Rizvi, Proc. of SPIE-IS&T Electronic Imaging, SPIE Vol. 7245, 72450K
© 2009 SPIE-IS&T · CCC code: 0277-786X/09/$18 · doi: 10.1117/12.805726

SPIE-IS&T/ Vol. 7245 72450K-1


two further problems arise. First of all, differently from JPEG coding of still pictures, recent video encoding
standards compute the DCT on variable-size blocks: 4×4 or 8×8 blocks are chosen according to the picture
motion and content. Secondarily, a video sequence contains independently coded frames (I frames) as well as
referenced frames (P and B frames). The latter are encoded as their difference with previous or subsequent I
frames, after motion compensation. If such I frames are affected by blocking artefacts, probably these artefacts
will be transmitted to referenced P and B frames but, because of motion compensation, they will be shifted to
different positions. Therefore, if a quality evaluation of such a P or B frame is needed, the correct position of
the blocks must be estimated. This problem can be ignored if the encoded video bitstream is available, so that
it is possible to identify I frames, and perform a quality evaluation on them only. However, if this evaluation
takes place after decoding, e.g. on a chip improving the quality in a monitor, a good quality evaluation is needed
for all frame types. At the best of our knowledge, no devised blockiness metric accounts for object motion.
Most blockiness methods assume regular blocking grid in a fixed position. Other methods, e.g.,10, 11 cope with a
block shifting, but still take a regular uniform grid for granted. A very clever estimation of the intensity of the
artefacts arising from the shifting of block boundaries is done in.12 However, it relies on the difference between
DCT coefficients of corresponding blocks in referenced and independent frames. Therefore, in the first place, we
do not consider it a proper blockiness metric, since it operates on whole blocks indistinctly, and not specifically
at block boundaries. In the second place, as stated, we assume to have access to the decoded video only, so
that a subsequent Discrete Cosine Transform computation would be needed. In some applications, especially
real time ones, this could be too expensive. We first adapt the method to generate synthetic blockiness to video
sequences. Then we present an original method to locate the blocks position.

2. BLOCKING ARTEFACTS IN JPEG COMPRESSION


2.1 Generation of synthetic blocking artefacts
In order to get quantitative results about the performance of the proposed metrics, it is expedient to be able
to generate synthetic blockiness artefacts, and suitable algorithms should be examined for this purpose. These
algorithms should satisfy a set of criteria; in particular, they should:

- generate pure artefacts, i.e. in our case they should yield only blockiness and not high-frequency errors or
blurriness or other distortions which may affect the measurement;
- be realistic, i.e. the generated artefacts should visually correspond to actual defects which can be seen in
real-world image representation;
- be computationally simple, in order to permit a real-time operation in their prospective usage in video
blockiness evaluation.

If limiting the computational load is mandative, a simple approach is to generate a set of blocks having random
luminance and to superimpose them on the image. The luminance of the various blocks can have a uniform
distribution in a range that increases with the strength of the desired artefact. Superposition may be additive,
with zero-mean blocks, or multiplicative, with blocks having an average value of one. In the latter case, of course,
the artefacts are stronger in bright portions of the scene; since it is known that the HVS has reduced sensitivity
to luminance changes where the average luminance is higher (Weber’s or Michelson’s formulas), this may affect
the measurement.
If realistic blockiness is mandative instead, one can resort to JPEG compression with a varying quality
factor. In the real world, indeed, JPEG compression is the main source of blockiness artefacts. It can be
objected however that in this way also other defects will be generated, negating the purity criterion. Purity of
the artefact is achieved with the method proposed in;13 it is an improved version of the approach described in
ITU-T Recommendation P.930.14 In this method, the average of each block in the picture is calculated, together
with the average of its adjacent blocks, and the difference of the two is calculated; this difference is scaled and
added to the original center block. Some corrections permit to avoid saturations and excessive blockiness.
A compromise among the mentioned criteria is proposed here. The blocking effect is achieved as follows: the
test image is divided into blocks, which are transformed with a 2-D DCT; the resulting coefficients are quantized

SPIE-IS&T/ Vol. 7245 72450K-2


with a table derived from the JPEG compression scheme, modified in order to affect only very low frequency
components; the inverse transform is eventually performed. The modified quantization matrix is obtained from
the standard JPEG matrix, keeping only its six lowest frequency terms while all the remaining ones are set to
one. The matrix is then changed according to the desired quality factor with the conventional algorithm:

if (Quality == 100); ScFact=1; end;


if (Quality < 100 & Quality > 50); ScFact = 200 - 2*Quality; end;
if (Quality <= 50 & Quality >= 1); ScFact = 5000 / Quality; end;
Q = round((ScFact.*Qdefault+50)/100);

Figure 1. Boat picture after JPEG compression (left) and after synthetic blockiness introduction (right)

Figure 1 shows a detail of the Boat image, after synthetic blocking artefact introduction (left) and after
JPEG compression (right), using the same quality factor Q=8 (a small value has been chosen in order to make
the result better visible). It is clearly seen that the image on the right has the same blocking appearance as the
one on the left, while the high-frequency artefacts that are typical of JPEG compression and that could affect
the subsequent blockiness measurement are not present. Noise is also visible in the left image, which is present
in the original data and is attenuated in the JPEG version.

2.2 Blockiness visibility

Human Contrast SensitMty


1000

0.1 10 100
log Background Luminance (cd/mA2)
Spatial Frequency (cycles/degree)

Figure 2. Contrast sensitivity function (left) and Threshold luminance (right).

SPIE-IS&T/ Vol. 7245 72450K-3


In order to obtain significant results, as mentioned above the blockiness metric should closely represent the
actual visibility of the blocks. Various aspects of the HVS response should be taken into account for this purpose;
the most relevant of them, at least until only still images are considered, are the Contrast Sensitivity Function
(CSF) and the brightness perception.
The CSF characterizes the response of the HVS to luminance changes with different spatial frequencies and
amplitudes. More precisely, it shows the minimum amplitude of a sinusoidal grating that can be perceived as a
function of the spatial frequency in cycles per degree. A plot of a typical photopic CSF is shown in Figure 2 at
left. Of course a practical use of this curve means that a hypothesis is made on the setting of the display/observer
system, and on some properties of the image which is visualized. As an example, for a modern 16:9 40” TV
observed from a distance of 1.65 m the horizontal angle of view is 30 deg, as recommended by the SMPTE.
If the display resolution is 1920×1080p, an 8×8 pixel block represents an angle of 0.22 deg, and the blocking
artefact in a uniform image area represents a square signal of period 0.5/0.22 = 2.3 cycles/deg, which is in the
high sensitivity portion of the CSF. If, however, the picture shown has an original lower resolution and has been
scaled before display, the size of the blocks can be significantly larger and correspond to a frequency at which
the HVS is less sensitive.
The CSF itself actually is a function of the luminance of the display on which the sinusoidal grating is shown.
A reference value of 100 cd/m2 can be taken, but if the light changes the behaviour is different: this may become
important in dark parts of the displayed image, where the Weber’s law is no longer valid and the threshold of
sensitivity shows a relative increase (see Figure 2).15 The ambient illumination has also an influence.

2.3 Blockiness measurement


A technique for the measurement of the blockiness artefact is suggested in.8 The image under test is first
subsampled, keeping only the samples at the four corners of each block (actually, the second row of the block
is used instead of the first one, to make the method suitable also for images derived from an interlaced source);
then, operating in the transform domain, the phase correlation is measured between selected couples of pixel
positions, in order to compare couples which are located inside the same block (intra) or across the borders of
two adjacent blocks (inter). The maxima of such correlations are added separately for intra and inter couples,
and the ratio of the two summations is the blockiness measure. Indeed, if no blockiness is present there is
no difference between intra and inter couples, and the output is one; for blocky images, on the contrary, the
inter-block correlation will in principle be smaller, and the ratio will increase.
The method was later improved in.9 More samples of the original images are taken in this case, since the
subsampling includes all the image rows and columns which form the contours of the blocks (again, the row
positions are moved to permit to measure also deinterlaced pictures). This makes the method more sensitive also
to situations in which only one of the borders of the blocking structure is visible. The two contributions from
the rows and columns are kept separate until their intra/inter ratio is calculated; then the difference between the
value two and the sum of the two ratios is calculated, and constitutes the output of the measure. Like above,
each ratio will be close to one when no blockiness is present, so that a measured value of zero will indicate a
clean image.
A different metric is proposed here, still based on correlation. The image is first processed with a very mild
lowpass filter; this refrains high-frequency components of the image noise from affecting the measurement. The
image, supposed to be formed by nr rows and nc columns, is then downsampled in order to form four vectors,
v3 , v4 , v8 , v9 , of size(nr × nc, 1): v3 is the concatenation of the image columns located in the positions 8n + 3,
where n = 0 → nc 8 − 1; similarly for v4 , in positions 8n + 4; and for v8 and v9 , in 8n + 8 and 8n + 9 respectively.
The Pearson’s correlations between v3 and v4 and between v8 and v9 are taken in the data domain, and their
scaled and shifted ratio is the blockiness index. With respect to to the method in,9 the proposed one has the
following advantages:

- it is computationally simpler, since the DFT evaluation is not needed;

- it avoids the need of introducing tweakings in order to cope with both interlaced and progressive data;

SPIE-IS&T/ Vol. 7245 72450K-4


- it is able to check the whole image area, while8 and9 use a Hamming window which effectively reduces the
examined area to the central portion of the image

The proposed method has proven to be able to sense the presence and the intensity of different types of
synthetic blocks and, more importantly, to quantify the blockiness introduced by the JPEG compression, giving
output values which monotonously increase for decreasing values of the quality factor used in the JPEG algorithm.
By measuring the artefact intensity of a JPEG compressed image and of an image with synthetic blocking
added using the above described modified-quantization-table method, we can also verify the robustness of the
proposed technique to other artefacts: the output of the measurement system should be the same on the two
images, for different values of the quality factor. Some first experiments, whose results are shown in the following,
indicate that this is a difficult issue. It can be seen however that the robustness can be increased by a suitable
data preprocessing, which has also the effect of making the metric closer to human perception. Two techniques
are proposed:

- a log mapping can be applied to the image, which complies with Weber’s law16 and makes dark areas more
important than bright ones;
- smooth areas operation: strongly textured and detailed portions of the image, where blockiness is percep-
tually less important, can be excluded from the measurements. For the latter purpose, we just operate
along a rasterized column vector of image data; it is first lowpass filtered, and the absolute values of the
difference between the original and the filtered data is evaluated. If this difference is larger than a threshold
(which is a fraction of the s.d. of the data), the sample is discarded.

2.4 Experiments on blockiness measurement


Experiments with the three different metrics (without preprocessing, with log mapping, keeping only smooth
areas) show that with respect to the no-preprocessing case the log mapping makes the metric less sensitive to
small blocking effect amounts (the output values are smaller for high Q factors), while the smooth areas selection
has the opposite effect. Subjective testing is needed in order to understand which of these approaches most closely
represents the HVS perception of the blockiness artefact. We show the results obtained on two test images with
quite different content, Lena and Boat. Each image is corrupted in two ways, i.e. by introducing blockiness
using the technique described above and by compressing it with JPEG. The amount of corruption is controlled
in both cases using the JPEG quality factor Q, in the range [5,100]. The figure below shows measurement results
on Lena (left) and on Boat (right); ’blj’ is the plot obtained from the JPEG compressed image, ’blq’ is the one
from the image with synthetic blocking.

250
200
blj
blj
blq
blq 200 -

150
150-

100 100-

50-
E
50
0-

-50
20 40 60 80 100 20 40 60 80 100
JPEG qualityfactor JPEG qualiti factor

Figure 3. Measured blockiness of Lena (left) and boat (right). blj: real JPEG compression blq: synthetic blocking

If the smooth areas operation option is activated, one obtains the two plots below. Since the ’blj’ and ’blq’
plots are closer one to each other, we can say that this metric is more robust to artefacts other than the blocking
effect. The price to pay is a reduced uniformity in the measurements, which are no longer monotonous with
respect to the quality factor.

SPIE-IS&T/ Vol. 7245 72450K-5


180

160
C0
160
a)
140
140

120 120

100
100 63
0
E
80

20 40 60 80 100 20 40 60 80 100
JPEG qualityfactor JPEG qualityfactor

Figure 4. Measured blockiness, smooth area option activated. Lena (left) and boat (right). blj: real JPEG compression;
blq: synthetic blocking

1200
blj
1000 -- - - blo enh
blo urn
800
a,

600
AD
13
400

200

200
0 20 40 60 80 100
JPEG qualityfactor

Figure 5. Measured blockiness after enhancement. blj: real JPEG compression; blo enh: coring applied; blo um: unsharp
masking applied

2.4.1 Measuring the effects of image enhancement operators on image blockiness


The proposed metric is very convenient if it is desired to estimate the amount of blockiness that is perceived
after having applied a sharpening operator. This is illustrated in the following experiment. The Lena image is
JPEG coded with Q in the range [5,100]; the resulting image is processed with two detail enhancement operators:
coring,17 and conventional unsharp masking. The measured blockiness in the three cases is plotted in Figure 5:
’blj’, ’blo enh’, and ’blo um’ respectively. It is seen that of course the blockiness measure yields increased values
after enhancing and even more after unsharp masking. When the differences between before and after processing
are significant, a blocking artefact amplification has been detected. Sample images at Q=60, 80, and 95 are
reported below. Notice that for Q=80 blockiness is barely visible in the image before the enhancement, but it is
visible after coring and after unsharp masking (and indeed the measured values are 9, 32, and 47 respectively).
For Q=90 blockiness is absent both before and after enhancement, and indeed the measured values are 8, 10,
and -2 respectively. Extremely small measured values, or even negative values like the last one, are caused by
high-frequency noise in the absence of blockiness. Based on these measured values, one could automatically
disable the sharpening, or modify its parameters. or even introduce a further deblocking stage, if the difference
in measured blockiness is above a certain threshold value.

SPIE-IS&T/ Vol. 7245 72450K-6


Figure 6. From top to bottom: JPEG images with Q=60, 80, 90; from left to right: no processing, coring, unsharp masking

3. BLOCKINESS IN VIDEO SEQUENCES


3.1 Artefact generation
To isolate the problem of the shifting of the block discontinuities in referenced frames, we need a sequence whose
P frames are only affected by a propagation of the artefact from the reference I frame, and do not add new
blocks. We want therefore to perturbate I frames with a regular blocking grid, and have this grid altered by
motion conpensation in subsequent P frames. On this purpose we take an original uncorrupted .avi sequence,
we introduce a synthetic blocking grid on every 12th frame, and then we encode this artificial sequence with an
H264 encoder, where the Group of Picture size is set to 12. This way we are sure that the artificial artefact is
on I frames only. To preserve our artificial artefacts from the interference of actual artefacts generated by the
encoder, we have to reach a compromise. On one side, in fact, we want to keep the encoding bit rate low, to avoid
that in the encoding of P frames enough bits are available to perfectly represent the difference with the corrupted
reference frame, therefore annihilating the synthetic artefact. On the other side, however, we want to keep the
bitrate high enough to avoid that real artefacts are added to synthetic ones in I frames. Since empirically we see
that I frames are much more affected by rate reduction than P frames, our compromise is a bitrate of 1 Kbps
and a value of 4 for the ratio of the qualities of P and I frames. These values may need to be modified according
to the selected test sequence.

3.2 Block detection


Our detection method is based on the behaviour of the luminance of the pixels in a row or a column of a
blocking–affected picture. We illustrate the detection of vertical block boundaries through the examination of a
pixel row, but the same procedure is applied to columns.
First of all, blocks are visible when they are abrupt discontinuities between pairs of pixels in an otherwise

SPIE-IS&T/ Vol. 7245 72450K-7


smooth line. We choose therefore to consider a local maximum of the pixel difference as a candidate for a block
discontinuity location, i.e.
Ic = {j | d(j) > d(j − 1) ∧ d(j) > d(j + 1)} (1)

 
d(j) ≡ I(i, j + 1) − I(i, j) for any row of index i.
Among this set of candidates, we take as indices of block borders those satisfying some further conditions:

1. To be detected, a block discontinuity has to be large enough. The first condition is therefore:

C1 : d(j) > T h1mindisc (2)

2. Blocks are most visible in generally uniform areas. The difference between pixel pairs adjacent to a block
border is usually small. This can be expected at least on one side of the discontinuity. We set therefore:
2

C2 : min d(id + s · j) < T h2maxsteep (3)
{s=−1,1}
j=1

id ∈ Ic , meaning that at least one of the sums of the two preceding and the two subsequent differences
must be below a threshold.
3. In a line section containing a smooth edge, or another gradual intensity variation, a block border could be
detected if its discontinuity overwhelms those of the natural variation, i.e. if the following statement is true:

d(id )
C3 : > T h3minratio (4)
min{j=−1,1} d(id + j)

4. For a pixel on a block border, the difference with its neighbour across the border  is much larger  than
that with its neighbour along the block boundary, i.e., if we define d (i, j) = I(i, j + 1) − I(i, j) and
  h
dv (i, j) = I(i + 1, j) − I(i, j), then, in order for a candidate jd ∈ Ic to be chosen as a block discontinuity
location, the following condition must hold:

dv (i, jd )
C4 : < T h4maxangle (5)
dh (i, jd )

We decide that a candidate (local maximum) is a block border if the following combination of the former
conditions
 is true:

C1 ∧ (C2 ∨ C3 ) ∧ C4
meaning that a local maximum in the discontinuity is deemed to be a block border if it is large enough and
lying inside a smooth area, or if it is inside an area of varying intensity, but it is much larger than this natural
variation. Further, an essential condition is that the cross-block variation is much larger than the variation along
the block border.
This method detects block discontinuities in each line (row or column) independently, without exploiting the
correlation among lines crossed by a block boundary. In fact, even if one condition, for example smoothness or
minimum discontinuity, does not hold for a local maximum in one line, a blocking artefact could however be
visible in that position, because the condition holds for all neighbouring lines, and the eye fills the gap. This
circumstance is exploited in the extended method proposed in the following.

• One line is taken at a time, to detect its block discontinuities. The local maxima of its differences are found
like in the previous example.
• In the found locations not only the current line but also its closest neighbours (for example, (N-1)/2 lines
above and (N − 1)/2 lines below) are examined.

SPIE-IS&T/ Vol. 7245 72450K-8


• Left side members of conditions C1 - C4 are computed for each line, summed over lines, and compared with
the corresponding thresholds multiplied by the number of lines N.

This way, if a line falls short to satisfy a condition, this can be compensated for by its neighbours.

3.3 Results

Figure 7. Detail of a frame affected by blockiness (above) and detected block discontinuities (below)

The detected bocking artefact positions are shown in Figure 7, superimposed on the frame. It can be seen
that, with the exception of too complex or too light block borders, the image blocks are effectively detected.
Once the block boundaries are individuated, any of the methods proposed so far to quantify the blockiness
intensity can be used.

4. CONCLUSION
The problem of measuring the intensity of the blocking artefact was addressed in this paper. In the first part the
artefact was measured in still pictures. First a way was found to generate a synthetic and pure artefact, then a
new efficient detection strategy was introduced and measurements were performed on naturally and artificially
affected pictures. The two main requirements for the method, i.e. robustness to other artefacts and ability to
follow the compression factor monotonically, were found slightly difficult to pursue at the same time, but were
effectively achieved individually. Besides, the measurement properly detects the effects of image enhancement
on blockiness. In the second part, the problem of block detection in video frames was addressed. This problem
is typical of video, where motion compensation intervenes, so that the position of the blocks is unknown. The
proposed method quite effectively estimates this position, and enables a more efficient application of existing
measurements methods which assume it to be known.

5. ACKNOWLEDGEMENTS
The second author gratefully acknowledges the financial and technical support of Philips Consumer Electronic,
Eindhoven.

REFERENCES
[1] Yu, Z., Wu, H., Winkler, S., and Chen, T., “Vision-model-based impairment metric to evaluate blocking
artifacts in digital video,” in [Proceedings of the IEEE ], 90, 154–169 (2002).
[2] Triantafyllidis, G., Tzovaras, D., and Strintzis, M., “Blocking artifact detection and reduction in compressed
data,” IEEE Trans. on Circuits and Systems for Video Techn. 12, 877–890 (2002).

SPIE-IS&T/ Vol. 7245 72450K-9


[3] Petrovski, A., Kartalov, T., Ivanovski, Z., and Panovski, L., “Blind measurement and reduction of blocking
artifacts,” in [48th International Symposium ELMAR, Zadar, Croatia ], 73–76 (2006).
[4] Ye, S., Sun, Q., and Chang, E.-C., “Detecting digital image forgeries by measuring inconsistencies of blocking
artifact,” in [Proc. of the IEEE international conference on multimedia and expo], 12–15 (2007).
[5] Hu, H. and de Haan, G., “Simultaneous coding artifact reduction and sharpness enhancement,” in [Proc.
of the International Conference on Consumer Electronics, 2007. ICCE 2007. Digest of Technical Papers.],
1–2 (2007).
[6] Wang, Z., Bovik, A., and Evan, B., “Blind measurement of blocking artifacts in images,” in [Proc. of the
IEEE International Conference on Image Processing], 3, 981–984 (2000).
[7] Wu, H. and Yuen, M., “A generalized block-edge impairment metric for video coding,” IEEE Signal Pro-
cessing Letters. 4, 317–320 (1997).
[8] Vlachos, T., “Detection of blocking artifacts in compressed video,” Electronics Letters. 36, 1106–1108 (2000).
[9] Farias, M. and Mitra, S., “No-reference video quality metric based on artifact measurements,” in [Proc. of
the IEEE International Conference on Image Processing], 3, 141–144 (2005).
[10] Shao, L., Wang, J., Kirenko, I., and de Haan, G., “Quality adaptive trained filters for compression artifacts
removal,” in [Proc. of the IEEE International Conference on Acoustics, Speech and Signal Processing],
897–900 (2008).
[11] Muijs, R. and Kirenko, I., “A no reference blocking artifact measure for adaptive video processing,” in [Proc.
of the European Signal Processing Conference], (2005).
[12] Leontaris, A., Cosman, P. C., and Reibman, A. R., “Quality evaluation of motion-compensated edge artifacts
in compressed video,” IEEE Trans. on Image Processing 16, 943–956 (2007).
[13] Farias, M., Foley, J., and Mitra, S., “Detectability and annoyance of synthetic blocky, blurry, noisy, and
ringing artifacts,” IEEE Transaction on Signal Processing 55, 2954–2964 (2007).
[14] International Telecommunication Union, Geneva, Switzerland, ITU-T Recommendation P.930: Principles
of a reference impairment system for video (1996).
[15] Ferwerda, J., Pattanaik, S., Shirley, P., and Greenberg, D., “A model of visual adaptation for realistic image
synthesis,” ACM Transactions on Graphics (SIGGRAPH 96) , 249–258 (1996).
[16] Guarnieri, G., Carrato, S., and Ramponi, G., “Nonlinear mapping for dynamic range compression in digital
images,” in [Proc. of the IS&T/SPIE Symp. on Electronic Imaging: Image Processing: Algorithms and
Systems VI], Loew, M. H., ed. (2007).
[17] de Haan, G., [Digital Video Post Processing], Royal Philips Electronics N.V., Eindhoven, The Netherlands.
(2006).

SPIE-IS&T/ Vol. 7245 72450K-10

View publication stats

You might also like